Include available ordinary captives, captured foreign gholas and shared Duke in the original random pool. Apply one public ordinary death without execution income, preserve foreign revival custody and other concealed captives, and end Duke tenure. Find the signed Semuta victim by canonical physical identity after captivity ends. Explain the provisional normal-Tanks source composition in controls and developer reference.
Validation:9 files74/74, types/lint/build and two actual-rule/rendered-table programmes plus four finite policy continuations pass on one frozen source. Details and exact unchanged-source proof are recorded in the canonical Residual Poison contract. Genuine public Advanced cache purchase, funded deployments, native victory/capture/keep and second battle cover native and captive deaths plus finite completion. Four minimal profiles use legal offered card/readiness actions. Separate controlled cases cover ghola/Duke, last-native return and Semuta claim; no strategic AI refinement.
This interpretation remains provisional. Wider interactions, complete modes, final assurance and browser/deployed acceptance remain open. No saved-room/database/reset/restart/deployment operation.
Replace the Basic-then-flip battle fixture with actual Advanced setup and add
five tests covering Truthtrance, Voice, ordinary Prescience, Ghola preparation,
private inspection and interrupted decisions. No runtime changes were needed.
Keep the special offer deadline and target-first ordering explicitly provisional.
Update the development maintenance policy per the user's instruction: restart
only when a change or observed server condition warrants it, preserving saves
and verifying restoration. No hourly schedule or recurring automation.
Validation: npm run check passed: types, lint and all 4,205 offline tests
(124.980 seconds). Source fingerprint:
5792403ddfed7ea7f4e0b247d367e7c000dfce6c81e1553204d9c00db376c3f6.
Private report: 20260913-full-plan-final/report.json.
Earlier build and HTTP evidence are reused for unchanged runtime code.
Independent source/test review found no actionable defect in this boundary.
Browser AZC4HGLB covers human inspection and Medium AI Ghola preparation,
desktop/phone readability and private refresh after the completed restart.
The earlier inventory seat also restored; all 64 saved room versions/hashes
remain unchanged. No mode or publication gate is opened.
Aggregate goal counters: 59,151,671 at batch start to 59,509,237 after completed
verification, or 357,566 tokens before final Git and new goal-steering overhead.
One bounded workhorse helper reused for source and independent test review;
one stable broad check. This is an audit batch, not a new gameplay capability.
Quintine